Série des Documents de Travail

Transcription

1 Série des Documents de Travail n The Effect of Public Policies on Consumers Preferences : Lessons from the French Automobile Market X. D HAULTFOEUILLE 1 I. DURRMEYER 2 Ph. FÉVRIER 3 October 2013 Les documents de travail ne reflètent pas la position du CREST et n'engagent que leurs auteurs. Working papers do not reflect the position of CREST but only the views of the authors. 1 CREST. xavier.dhaultfoeuille@ensae.fr 2 Corresponding author : University of Mannheim. isis.durrmeyer@ensae.fr 3 CREST. philippe.fevrier@ensae.fr

2 The Effect of Public Policies on Consumers Preferences: Lessons from the French Automobile Market Xavier D Haultfœuille Isis Durrmeyer Philippe Février September 2013 Abstract In this paper, we investigate whether French consumers have modified their preferences towards environmentally-friendly vehicles between 2003 and We estimate a model of demand for automobiles incorporating both consumers heterogeneity and CO 2 emissions of the vehicles. Our results show that there has been a shift in preferences towards low-emitting cars, with an average increase of 367 euros of the willingness to pay for a reduction of 10 grams of carbon dioxide per kilometer. We also stress a large heterogeneity in the evolution of preferences between consumers. Rich and young people are more sensitive to environmental issues, and our results are in line with votes for the green party at the presidential elections. We relate these changes with two environmental policies that were introduced at these times, namely the obligation of indicating energy labels by the end of 2005 and a feebate based on CO 2 emissions of new vehicles in Our results suggest that such policies have been efficient tools to shift consumers utility towards environmentally-friendly goods, the shift in preferences accounting for 20% of the overall decrease in average CO 2 emissions of new cars on the period. Keywords: environmental policy, consumers preferences, CO 2 emissions, automobiles. JEL codes: D12, H23, L62, Q51. We would like to thank Pierre-Louis Debar and Julien Mollet from the CCFA for providing us with the data. We also thank Sylvain Chabé-Ferret, Pierre Dubois, Philippe Gagnepain, Christian Huse, Iris Kesternich, Joshua Linn, Laurent Linnemer, Frank Verboven, Philipp Schmidt-Dengler, Katheline Schubert and participants of the various seminars and conferences for their remarks. CREST. xavier.dhaultfoeuille@ensae.fr University of Mannheim. Corresponding author. isis.durrmeyer@ensae.fr CREST. philippe.fevrier@ensae.fr

3 1 Introduction In this paper, we study how people reacted to two French environmental policies that aim at mitigating automobiles carbon dioxide (CO 2 ) emissions. The first is the implementation, at the end of 2005, of a European directive compelling manufacturers to indicate CO 2 emissions for every car. The second is the introduction, in January 2008, of a green taxation called the bonus/malus system (referred to as feebate hereafter), which provides a financial reward for low CO 2 emitting vehicles (less than 130 grams per kilometer) and a penalty for the most polluting ones (more than 160 grams per kilometer). More precisely, we investigate whether French consumers have modified their preferences towards environmentally-friendly vehicles between 2003 and The first reason for this interest is an environmental concern. In the last two decades, environment, and in particular global warming, has become a major issue. Policy initiatives are launched in many countries to reduce the human contribution to greenhouse gas emissions, especially CO 2. Cutting vehicle emissions is a crucial objective, as the transportation sector accounts for a third of the CO 2 emissions in developed countries. In April 2010, 17 European countries have implemented a taxation related to the average CO 2 emissions of the vehicles. 1 The California Clean Cars Law, introduced by the State of California and followed by 13 other States, is another example. This program has the ambition to reduce overall greenhouse gas emissions from passenger cars by 18% in 2020 and 27% in However, it is unclear how this growing concern for global warming at the society level translates at the individual one, both in terms of utilities and choices. First, global warming is a very slow phenomenon that will impact consumers in the long run only. Second, it is somehow immaterial and individuals may not know exactly what is their true individual impact on it. Finally, even if it enters in the utility function of the consumers, environment is a public good with a very large number of individuals contributing to it. Because of this classic free riding problem, people may not modify their choices, even if global warming and environmental issues are more and more discussed. At the end, we may thus wonder if the evolutions observed at an aggregate level correspond to better information and a true change in preferences, or just standard reactions to supply shocks stemming from environmentally-friendly technical changes and to new incentives created by public policies. The second reason to investigate the effect of these policies is related to the more general issue of how consumers react to public policies. Beyond incentive effects, public policies may 1 For recent analyzes of the environmental effect of such policies, see for instance D Haultfoeuille et al. (2013) and Huse & Lucinda (2013). 1

4 affect social preferences, which in turn modify individual behaviors. A growing economic literature, either based on theory, experiments or natural experiments, acknowledges the importance of such effects (see, e.g., Bowles & Polanía-Reyes, 2012, for a recent survey). Public policies may also modify the information set of bounded rational consumers, which, in turn, may affect their choices. One goal of the paper is thus to investigate whether such effects are at stake here, and, if so, to assess their importance with respect to more standard price and supply-side effects. To answer these questions, we use a dataset from the association of French automobile manufacturers (CCFA) that records all registrations of new cars in France between 2003 and 2008, as well as some individual characteristics of the purchasers. Compared to most of the existing literature that deals with the measure of environmental preferences, using such data presents two main advantages. First, we observe true choices as opposed to stated preferences, thus avoiding the so-called hypothetical bias (Arrow et al., 1993). Second, it is instructive to see whether environmental concerns matter in carefully thought purchases that represent a large share of consumers budget. We investigate, through a structural approach, how consumers preferences for CO 2 emissions and their willingness to pay to reduce global warming have evolved over this period of time. We also study whether this evolution is heterogeneous among consumers. We estimate a nested logit incorporating observed heterogeneity through 18 demographic groups of consumers based on age, income and urban area. Our findings are the following. We observe that the introduction of both policies coincides with a significant decrease of respectively 1.8 and 7.3 grams in the average CO 2 emissions of new vehicles, the average CO 2 emissions being around 155 grams per kilometer in An analysis of the market shares of each class of energy confirms these results. The market share of cars emitting between 100 and 120 grams of CO 2 per kilometer (class B), for instance, increased from 19.9% in 2007 to 38.4% in 2008, following the introduction of the feebate. Conversely, the market share of cars emitting between 160 and 200 grams of CO 2 per kilometer (class E) has sharply decreased from 18.0% to 9.9%. Disentangling between pure price effects of the feebate, changes in preferences of the consumers and other effects, 2 we find a coincidence between the evolution of the consumers utility and the timing of the implementation of both policies. Our results thus suggest that environmental policies have been efficient tools to shift consumers utility towards environmental friendly goods. We obtain that between 2003 and 2008, CO 2 emissions have been reduced by more than 10%. 2 Changes in preferences should be understood in a broad sense, including informational effects of the policies. 2

5 20% of this decrease are related to the evolution of consumers preferences, 51% stem from the price effect of the feebate while 29% account for supply-side and other effects. We thus find evidence that consumers value environment and the reduction of global warming, and that their valuations has increased over time. This is true for all the consumers we are considering, though we find a substantial heterogeneity in this evolution. It differs in particular along age and income, the youngest and the richest being those who value the most the environment. Between 2003 and 2008, young consumers increased their valuation of the reduction of global warming twice more than old ones, while rich consumers increased their valuation around 1.5 times more than poor ones. In line with this interpretation, we observe a positive correlation between the average evolution of the willingness to pay at thex town level and the result of the green party candidate at the 2007 presidential election. Finally, combining the estimates of environmental preferences with price elasticities, the willingness to pay for a reduction of 10 grams of CO 2 per kilometer raised on average by 367 euros in 2008 compared to These results are consistent with Brownstone et al. (2000) and the results of the MIT Survey of Public Attitudes on Energy and the Environment. The paper is organized as follows. Section 2 presents the environmental policies, the evolution of average CO 2 emissions on the period and potential explanations for this evolution. Section 3 presents the demand model and the results. Section 4 concludes. 2 Environmental policies and evolution of CO 2 emissions 2.1 Energy labels and the feebate system By the end of 2005, the European Commission compels manufacturers to place an energy label on each new car. The French decree applying this European directive was published in November 2005 and manufacturers were given six month, i.e. until May 2006, to conform to it. The policy still applies today. The energy label indicates the precise average CO 2 emissions of the vehicle and fuel consumption (in liter for 100 kilometers), its class of emissions and the position of this class among all classes (see Figure 1). Seven classes are defined, from A, corresponding to the lowest CO 2 emitting cars (less than 100 grams per kilometer), to G, the highest emitting ones (over 250 grams per kilometer). The goal of this policy is to encourage consumers to buy greener cars by informing them about CO 2 emissions. Thanks to these energy labels, consumers are aware of the impact of each car 3

6 on global warming. They may thus take it into account in their purchase decision, whereas it was more difficult to do so before the policy. This informational aspect is reinforced by the choice of the colors associated with the classes: from green for class A to red for class G. These colors were deliberately chosen to influence consumers and signal them what a good purchase for environment is. Figure 1: A model of French energy label. The second institutional change is the introduction, in January 2008, of a green taxation called the bonus/malus system, referred to as feebate hereafter. This new policy was announced on October 25, It was one of the main measures of an environmental roundtable called the Grenelle de l environnement that took place in France in Its purpose, among others, was to lower average CO 2 emissions stemming from cars from 176g to 130g of CO 2 per kilometer in 2020, and the feebate was chosen as an incentive to purchase environmentally-friendly new vehicles. To this end, a financial rebate, from 200 and 1,000 euros, was given to consumers who buy low CO 2 emissions level vehicles (less than 130g/km), while consumers buying polluting cars (more than 160g/km) were taxed, from 200 to 2,600 euros. The exact amount of the rebate or the fee depended on the class of emissions the vehicle belongs to and the entire scheme is presented in Table 1. These classes correspond to those of the energy label, in 4

7 which the subclasses C+, C-, E+ and E- were introduced. 3 This feebate is received or paid once, at the time of the sale of the vehicle. It applies to all new cars, including those purchased abroad. On the other hand, second-hand vehicles were not in the scope of the policy. Class of Emissions Feebate Percentage of emissions (in g/km) 2007 prices A (60-100] +1000e 8.1% B ( ] +700e 4.8% C+ ( ] +200e 1.2% C- ( ] 0e 0.0% D ( ] 0e 0.0% E+ ( ] -200e -0.98% E- ( ] -750e -3.2% F ( ] -1600e -4.3% G > e -5.2% Table 1: Details of the feebate Contrary to the first policy, which only aims at modifying the information given to the consumers, the feebate policy introduces financial incentives to encourage them to buy an environmentally-friendly vehicle. These incentives are important in magnitude, the rebate representing up to 8.1% of the list price on average for class A, and the penalty rising to as much as 5.2% of the list price for class G. 2.2 Evolution of CO 2 emissions Before decomposing finely the effects, we provide a broad picture on the evolution of average CO 2 emissions of new cars in France. We rely for that purpose on a dataset provided by the Association of French Automobile Manufacturers (CCFA, Comité des Constructeurs Français d Automobiles), which records all the registrations of new cars 3 We do not indicate in this table the class of emissions A+, which corresponds to emissions lower than 60g per kilometer. A rebate of 5000e was associated to this class, but in 2008 no vehicle belonging to this class was sold in France. Note also that for the replacement of more than 15-year old vehicles by new vehicles, the rebates were increased by an amount of 300 euros. This only represents a very small fraction of the total amount of rebates (2.6%), and we neglect this measure hereafter as we do not observe which purchaser received this extra rebate. 5

8 bought by households from January 2003 to January Figure 2 displays the evolution of average CO 2 emissions of cars purchased on that period. Overall, there was an important reduction of 13% (from 156 to 136 grams per kilometer) on average CO 2 emissions of new cars between January 2003 and January This reduction can be decomposed in three parts. Between January 2003 and October 2005, before the introduction of the compulsory energy label policy, average CO 2 emissions dropped from 156g to 152g. This negative trend indicates that there was already, before the policies, a tendency to reduce CO 2 emissions. Between the two policies, from November 2005 to October 2007, the decrease was slightly more important, the CO 2 emissions falling from 152 to 147. This is reinforced by the fact that this decrease took place in a shorter period of time (24 months instead of 34). We finally observe a large drop after the introduction of the feebate. In February 2008, CO 2 emissions were equal to 138 grams and reached 136 grams in December The feebate policy seems thus to have had a huge impact on CO 2 emissions. We also see a peak in the average emissions in December 2007, followed by a large drop. This is probably due to anticipation effects. The policy was announced by the end of October 2007, so that some households who planned to buy a high CO 2 emitting vehicle were able to anticipate their purchase in order to avoid the penalty Compulsory energy labels Feebate Average CO2 emissions Linear trend (with breaks) Figure 2: Seasonally adjusted monthly average CO 2 emissions of new cars. 4 We exclude from this dataset exotic cars such as Rolls-Royces and Maseratis as well as commercial models and vans like Renault Master, which respectively represent 0.09% and 0.21% of the purchases. 6

9 To further analyze these patterns and measure the impact of the introduction of both policies, we regress the CO 2 emissions on three trends (corresponding to the three periods previously described) and on two dummies indicating if each policy was active or not at this time, controlling for monthly seasonal effects. The results are given in Table 2 and represented in dashed lines in Figure 2. The econometric analysis confirms that the feebate policy had a significant and negative impact on CO 2 emissions but also indicates that the introduction of compulsory energy labels negatively and significantly decreased the level of emissions. Even if the effect of the first policy was smaller than the one of the feebate (-1.81 versus grams of CO 2 ), it seems that both informational and financial incentives are important to modify consumers choices. We also see that not only the level but also the trend in the decrease of CO 2 emission have been affected and strengthened by both policies. Parameters Intercept Time trend Dummy of being after 11/2005 Estimate (0.370) (0.011) 1.81 (0.430) Additional trend after 11/ Dummy of being after 01/2008 Additional trend after 01/2008 (0.024) 7.27 (0.535) (0.069) Nb of observations 71 Notes: monthly effects are included. December 2007 and January 2008 are dropped because of anticipation effects. Standard errors are heteroskedasticity-autocorrelation robust. Significance levels: 1%, 5%, 10%. Table 2: Regression of CO 2 emissions on time Given that both policies are based on the classes of emissions of new cars, similar patterns should be observed in the market shares of each class of emissions between 2003 and We display their evolutions in Table 3. Overall, the results are in line with the effects depicted in Figure 2. The market shares of low-emitting classes increase sharply during the period, while those of high-emitting classes fall drastically. Considering for instance class B, we observe in 2006 a first jump from 13.2% to approximately 18.7% in its market 7

10 shares, and then an even larger increase, from around 20% to 38.4%, in Conversely, considering class E-, we observe in 2006 a first fall from around 17% to approximately 15.2% in its market shares, and then an even larger decrease, from 15.1% to 7.8%, in This is confirmed by the econometric analysis on classes B and E (see Figure 6 and Table 18 in Appendix B). Class of Market shares Evolution emissions A 0.03% 0.02% 0.01% 0% 0.02% 0.06% 100% B 13.58% 13.22% 13.17% 18.72% 19.89% 38.40% 188% C+ 2.61% 5.20% 10.60% 12.35% 10.33% 9.53% 264% C % 18.55% 14.84% 15.63% 19.74% 18.57% 12% D 36.04% 34.16% 32.66% 28.77% 26.93% 21.67% -38% E+ 4.76% 5.45% 5.31% 3.38% 2.90% 2.04% -55% E % 16.48% 17.04% 15.20% 15.10% 7.82% -60% F 4.80% 5.16% 4.98% 4.63% 3.66% 1.38% -73% G 1.82% 1.76% 1.39% 1.30% 1.44% 0.54% -73% Notes: market shares exclude the outside option (i.e., not to buy any car) and thus sum to one here. Table 3: Evolution of market shares according to the emissions class Finally, we study the evolution of average CO 2 emissions according to some demographic characteristics. This is possible since the French new cars registrations dataset provides information on the car but also on its owner. We observe the age and the city in which the owner lives. Based on these characteristics, we created 18 groups of individuals based on their age classes (18-39, or 60 and more), geographical areas (cities of less than 20,000 inhabitants, called rural areas or more, called urban areas) and imputed income classes (0-22,000, 22,000-32,000 or more than 32,000). Details on the income imputation and market definitions are provided in Appendix A.1. These three variables were chosen because they turn out to have a large effect on purchases (see Table 13 in Appendix A.1). The evolution of average CO 2 emissions according to the type of area, age and income are displayed in Table 4. The table emphasizes consumers heterogeneity in the purchase of new cars. Young people, low-income people, and people living in rural areas tend to buy lower CO 2 emitting cars. Nevertheless, the evolution of CO 2 emissions supports our previous findings and is rather the same for all groups of consumers. In particular, each policy implies a significant decrease in the CO 2 emissions, with, as expected, a larger effect 8

11 for the feebate. Demographics Evolution Age [18-39] % Age [40-59] % Age [60-100] % Income < 22,000e % Income [22,000-32,000] % Income > 32,000e % Rural area % Urban area % Table 4: Evolution of average CO 2 emissions according to demographic characteristics 2.3 Potential explanations The evolution of average CO 2 emissions may be due to several factors that we now examine in details Price effects The more natural explanation of the sharp decrease in CO 2 emissions in 2008 is a price effect due to the feebate system. What is peculiar is that the amounts of fees and rebates are relatively low compared to their effect. For instance, a rebate of 700e given to class B vehicles represents, on average, a 4.8% reduction of price (see Table 1), but implies an increase by almost 100% of the market share of this class. Similarly, the decrease of 50% in the market share of class E- would be due to a 750e penalty, or a 3.2% raise of their prices. Because manufacturers adjusted their prices in reaction to this policy change, these percentage are actually upper bounds of the true changes in prices. Overall, it thus seems difficult to rationalize such variations of sales with usual price reactions solely. These important variations may however stem from threshold effects induced by the feebate. If many consumers choosing class C+ vehicles before 2008 were nearly indifferent between them and class B vehicles, they could massively switch to class B following the introduction of the feebate. The extent of this effect can only be assessed through a microeconomic demand model taking into account heterogeneity both of cars attributes and consumers preferences. We develop and estimate such a model in the following sections to assess precisely the importance of the price effect. 9

12 2.3.2 Supply side effects The evolution of CO 2 emissions may also be due to supply side effects, such as technical changes or new incentives to market low-emitting vehicles, because of a fuel price increase for instance. The sharp fall in 2008 could also be due to an immediate adjustment of the supply side. To assess the importance of these effects, we study how the set of vehicles offered by manufacturers evolved through time. Because we do not observe directly this set of vehicles, through brand lists for instance, we assume that a product is proposed to consumers at a given month if it is bought at least once before this month and after this month, or at least once during this month. We then compute average CO 2 emissions of the set of cars offered at each date, for all brands and French brands only. The results are depicted in Figure 3. Feuil1 g/km 200 Compulsory energy labels Feebate All brands French brands 150 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06 Jan-07 Jul-07 Jan-08 Jul-08 Notes: each month, average is taken over all ``supplied'' vehicles, namely vehicles sold once before and after the month, without weighting by their sales. Our construction of ``supplied'' vehicles imply that at the beginning or end of the period, only vehicles with enough sales are included. These vehicles tend to have lower CO 2 emissions. To avoid such boundary effects, we drop the first and last six months. Figure 3: Average emissions of supplied vehicles Overall, there has been a reduction of around 5.5% of the average CO 2 emissions of supplied cars on the period. This decrease is very regular over time and approximately identical for French and other manufacturers. Beyond technical change effects, this could partly be due to the fuel price increase over this period. The gasoline price increases on average by 6.3% 10

13 per year, well above the average inflation in France over this period (2.2%). Long term objectives such as Voluntary Agreements may have also played a role. Since the end of the 90 s, automobile manufacturers committed to reducing the level of CO 2 for passenger cars in the European Union, the latest target being an average of 130 g/km for Finally, the observed decrease may be a reaction to an increasing trend in preferences for low-emitting cars, leading manufacturers to develop their product line in favor of small and fuel efficient vehicles. On the other hand, it seems that there is no immediate change in the products offered in response to both policies. This may seem surprising, especially for the feebate, given the sharp changes observed in market shares of the different classes of emissions. However, there are several reasons for not observing an immediate adjustment of the supply side. First, the manufacturers incentives may not be that large, because in January 2008, the feebate policy was conducted in France only. Although taxes related to CO 2 emissions of vehicles exist in most other European countries, they do not display similar discontinuities at the emission classes level. The advantage of exploiting these thresholds for the French market only may thus not overcome the costs of developing specific models, especially for non French manufacturers. 5 Second, the feebate policy was announced only two months before its application, and the very quick implementation of the reform contrasts sharply with the time needed by manufacturers to improve fuel efficiency. It is usually thought to take several years to develop new technologies and incorporate them in new vehicles. Berry et al. (1993), for instance, observed a two-year shift between the increase in the fuel price following the first oil crisis and the corresponding technical innovations. Similarly, we do not observe any particular acceleration or changes between 2003 and 2008 in the number of patents on domains related with CO 2 emissions. Finally, even if horsepower, and thus CO 2 emissions, can be adjusted quickly, the modified vehicle must be certified before appearing on the market. This certification, together with the distribution of the new vehicles, typically takes several months Macroeconomic effects It is well documented that the automobile industry is sensitive to macroeconomic shocks (see, e.g., Bar-Ilan & Blinder, 1992, Hassler, 2001). Even microeconomic studies put forward the importance of aggregate shocks (see, e.g., Goldberg, 1995). Negative economic conditions may lead people to buy smaller, low-emitting cars. It seems unlikely, however, 5 Note also that within countries tax systems evolve rapidly. The feebate cutoffs, for instance, were modified in 2010 and in This further limits the incentive to adapt the products line. 11

14 that the drop in January 2008 stems from such an effect. Figure 4 shows that the economic distress mainly appears at the fourth quarter of Similarly, the unemployment was at a historically low level in January 2008 (7.5%), and the sharp increase (from 8.1% to 9%) only occurs by the fourth quarter of Thus, consumers economic situation seems very unlikely to explain the pattern of CO 2 evolution. Yet, we do take into account possible wealth effects hereafter by stratifying our demand model along income classes and introducing year dummies to capture some of these effects. Fuel price is also known to be an important determinant of automobile choices (see, e.g., Berry et al., 1993). The increase after 2005 may be responsible for the change in the trend of average CO 2 emissions that we document before. It is very unlikely to explain the drop at the beginning of 2008, on the other hand, as the fuel price increase started at the end of Nonetheless, we take into account this evolution by including kilometers per euro in the attributes of the vehicles when estimating our demand model. 1,5 december december december december , , , Source : INSEE Source : INSEE Figure 4: Quarterly GDP Growth Figure 5: Gasoline Prices evolution Temporary effects Another possibility is that the large fall that occurred in 2008 may be due to temporary effects. First, consumers may anticipate the feebate to be temporary, and thus take advantage of the rebates quickly after the introduction. This kind of reactions would however be completely at odds with the government announcement. The feebate was supposed to be permanent, with only a decrease of the cutoffs by 5g every year from 2010, to induce technical progress. In practice, a change in the rebate amounts did occur in 2010, from 1,000, 700 and 200 euros for classes A, B and C+ to respectively 700, 500 and 0 euros. However, it seems unlikely that a rush in small cars purchase in 2008 could be due to the anticipation of this cut in the rebates. We would rather expect such a rush to occur by the end of

15 Second, price changes may imply a decrease in the optimal lifetime of smaller cars and an increase in the optimal lifetime of bigger ones. In this case, some individuals with small cars find it optimal to replace their car at the beginning of the period, while individuals with bigger cars postpone their replacement (see, e.g., Adda & Cooper, 2000, for evidence of such effects). If this effect was large, the fall in average CO 2 emissions should be quickly followed by a rise in these emissions. We do not observe such a rise in On the contrary, the trend in the decrease of CO 2 emissions is significantly higher after the beginning of Similarly, the market share of class B increases more quickly after this point. Even though we do not have monthly data in 2009, Table 5 shows, using aggregate data also from the CCFA, that this evolution continues in The market share of class A was multiplied by three between 2008 and 2009, while the one of class B increased by 36%. On the opposite, the market shares of classes E+ and G decreased by around 50%. Even though other phenomena are probably at stake in 2009, 7 these evolutions suggest that the sharp changes following the introduction of the feebate were not temporary. Class Market shares Evolution of emission of shares A 0.08% 0.29% +259% B 35.18% 47.8% +35.9% C+ 9.46% 7.99% -15.6% C % 17.1% -7.8% D 22.71% 17.94% -21.0% E+ 2.01% 1.07% -47,0% E- 8.98% 5.97% -33.6% F 2.27% 1.51% -33.6% G 0.74% 0.34% -54.5% Sources: 2008: detailed dataset on registrations of new cars (CCFA). 2009: aggregated data on registrations of new cars (CCFA). Table 5: Market shares according to the class of emission after 2008 (aggregated data) 6 For the sake of comparison, the 2008 figures include car fleets and some exotic cars that are excluded otherwise from our analysis. This explains why the market shares by classes differ from Table 3. 7 This year corresponds indeed to the peak of the economic crisis. The government also introduced a scrapping subsidy of 1,000 euros for more than 10-year-old cars that were replaced by vehicles emitting less than 160g/km. 13

16 2.3.5 Changes in consumers information and preferences Finally, additional non-price effects may be at stake. It is documented that people value environment per se, and are thus ready to pay for environmentally-friendly goods (on automobiles, see, e.g., Brownstone et al., 2000, or Potoglou & Kanaroglou, 2007). It seems plausible then that environmental policies shape and reinforce these preferences. Such changes in preferences would explain both drops at the end of 2005 and at the beginning of These policies may have modified the information set of consumers, by putting forward the CO 2 emissions levels of automobiles. The information being easier to incorporate in their choices, consumers may have taken it more easily into account. In the model developed by Gabaix (2012), consumers face too much characteristics and only select part of them to make their choices. If policies reduce the cost of gathering information about CO 2 emissions, consumers will rely more on this characteristic when purchasing a car. The feebate could also modify people s preferences through the informational content of the policy (see, e.g. Barigozzi & Villeneuve, 2006). Basically, the tax could be seen as a credible signal that environmental issues really matter, in a world where consumers may have trouble to make up their mind about the negative impact of CO 2 emissions. The introduction by the state of a tax, or a feebate as here, is a way to convince consumers that CO 2 emissions constitute a first order problem. Though we do not model this hereafter, the change in preferences may also be reinforced by peer effects. If people tend to conform to others in their purchasing decisions (see e.g. Bollinger & Gillingham, 2012, for such evidence), and anticipate that environmental policies will induce a shift towards environmentally-friendly cars, this shift will be reinforced through a social multiplier. All these effects will be captured in our model by a change in the valuations for CO 2 emissions. With a slight abuse of language, we will refer to them as changes in consumers preferences, keeping in mind that they could also be informational or peer effects. 3 Distinguishing the effects 3.1 The demand model To disentangle between the explanations detailed above, we rely hereafter on a structural demand model, using the CCFA registration dataset. CO 2 emissions, the brand, model, 14

17 type of fuel, number of doors, type of car-body, horsepower, weight and cylinder capacity are reported for each registration. These characteristics have been complemented with list prices 8 and fuel prices, allowing us to compute the number of kilometers per euro. We estimate a nested logit with observed heterogeneity, taking advantage of the availability of consumers characteristics in our database. In other words, we estimate structural parameters that are group-specific. This amounts to supposing that among each of the 18 groups of individuals defined by their age class, income class and type of area, preferences are homogenous. Formally, we suppose that each year, consumers can choose to buy one of the J different products proposed on the market. To avoid the aforementioned anticipation and postanticipation effects of 2007 and 2008, the year we consider actually excludes January and December. We define a product by its brand, model, car-body style, type of fuel, CO 2 emissions class and number of doors. As usually in the literature, we consider the characteristics of the base version of the car model, which is considered to be the cheapest one. Product 0 corresponds to the outside option, namely not purchasing a new car during the year. The automobile market is supposed to be segmented according to the main use of the car and we have created several nests accordingly. 9 The utility of consumer i, belonging to the demographic group d, for purchasing car j at year t is then given by p jt denotes the price of vehicle j at t. U d ijt = p jt β d + X jt γ d + f d t (CO 2jt ) + ξ d jt + ε ijt. X jt denotes other standard attributes: weight, horsepower, number of kilometers per euro (computed using the average price of fuel each year), engine capacity, number of doors and car-body style. X jt also includes time and model fixed effects, to control for macroeconomic shocks and unobserved heterogeneity of products, respectively. Thanks in particular to the introduction of the feebate, we still have sufficient variations within models, in particular in prices, to allow for such fixed effects. To capture potential time-varying environmental concerns of the consumers, we also include CO 2 emissions through the term f d t (CO 2jt ), where we consider several specifications for f d t hereafter. CO 2jt is not collinear with the number of kilometers per euro because first, this cost varies with the fuel type and second, gasoline and diesel prices differ. 10 Its specific effect can therefore be identified. If the environmental policies affects consumers utility, we 8 Transaction prices (including potential discounts by distributors) would be preferred, but are not available, as usually in this literature (see, e.g. Berry et al. (1995)). 9 Our segmentation is close to the one of the European New Car Assessment Program (Euro NCAP), see Appendix A.3 for details. For more details on the construction of products, see also Appendix A CO 2 emissions and kilometers per euro are related through the formula km/e = k CO 2 p g, where p g 15

18 should observe a change in the impact of CO 2 emissions in 2006, 2007 and 2008 compared to the previous years, all other things being equal. As explained above, the interpretation of the term ft d (CO 2jt ) is complex. It may capture both a negative valuation of CO 2 emissions per se and an information effect, people becoming more aware of the true CO 2 emissions of the cars and its impact on pollution level. ξjt d and ε ijt correspond to variables that are unobserved by the econometrician. ξjt d represents the mean valuation of unobserved attributes, such as the reliability or the design of the vehicle, for instance. Finally, ε ijt is the individual-product-specific error term. In the nested logit model, the (ε ijt ) j=1...j are allowed to be correlated for two vehicles in the same nest g. This takes into account the correlation in individual preferences for vehicles belonging to the same nest (family, executives, sports car...). The nested logit specification, together with the normalization to zero of the mean utility level of the outside option, 11 yields (see, e.g., Rust & Berkovec, 1985) ln(s d jt) ln(s d 0t) = p jt β d + X jt γ d + ft d (CO 2jt ) + σ ln( s d j/g t) + ξjt d, (1) where s d jt is the market share of product j and s d j/g t denotes the intra-segment share of product j among nest g. σ represents the correlation of consumers utility across automobiles of the same nest and lies between 0 (no correlation) and 1 (perfect correlation). This equation is very convenient for estimation because it provides a linear relationship between the market shares and the characteristics of the product. This equation also incorporates consumers heterogeneity through the dependence in d of β d, γ d, ft d and ξjt. d Hereafter we impose a linear decomposition in d, so that for the price parameter for instance: β d = β 0 + β u 1 urban + β a 1 1 age [40,59] + β a 2 1 age 60 + β i 1 1 income [22,000;32,000] + β i 2 1 income 32,000. As usually (see, e.g., Berry et al., 1995, Nevo, 2000, and Nevo, 2001), we suppose that, except prices, all characteristics are predetermined and uncorrelated with the error term ξ d jt. On the contrary, prices are allowed to be endogenous. This is typically the case if manufacturers observe the (ξ d jt) d,j and take them into account in their pricing strategy. 12 By construction, conditional market shares s j/g t are also endogenous, so that at least stands for the gasoline price and k depends on the fuel type only (k = if the motor uses gasoline, k = if the motor uses diesel). 11 The mean utility of the outside option may evolve through time, especially if the feebate policy has discouraged consumers to purchase cars on the second hand market. This is accounted for by the year dummies. 12 Another source of endogeneity is measurement error since, as mentioned before, we observe list prices rather than transaction prices. 16

19 two instruments are necessary to identify the demand model. Following the literature (see, e.g., Berry et al., 1995), our instruments are based on the characteristics of other products. If firms compete in prices on an oligopolistic market with differentiated products, they are constrained in their pricing strategy by the existence of close substitutes. The characteristics of the other products are thus likely to affect all prices, but are not correlated with the unobserved demand term ξjt. d Following this logic, we rely hereafter on four sets of instrumental variables. The first is the sums of characteristics of other brands products. The second is the sums of characteristics of other brands products of the same segment. The third consists of the sums of characteristics of other models of the brand. The last set is composed by the sums of characteristics of other models of the brand in the same segment. 3.2 Consumers preferences We estimate Model (1) as explained above but with slight variations in the way the price and CO 2 affect the model. In Specification (1), price is not instrumented, whereas all other specifications allow for price endogeneity. Specifications (2) to (4) differ in the way CO 2 emissions are included in the regressions. In Specification (2), the evolution of CO 2 preferences are captured through a temporal trend. In Specification (3), CO 2 emissions are interacted with year dummies whereas CO 2 emissions are interacted with two periods ( and 2008) dummies in Specification (4). Results are displayed in Table 6. For the sake of concision, we only present the estimates of σ and ( β, γ), the average of the preferences parameters (β d, γ d ) d= among the population of purchasers. Table 19 in Appendix B displays the estimates of the preference parameters according to demographic characteristics. 17

20 Variable (1) (2) (3) (4) Price ( β) (0.001) ln( s) (σ) (0.008) Characteristics ( γ) (0.012) (0.013) (0.013) (0.009) (0.008) (0.008) Weight (0.006) (0.017) (0.019) (0.019) Horsepower (0.004) (0.017) (0.018) Km/e (0.002) (0.004) (0.004) Cylinder capacity (0.002) (0.002) (0.002) Station wagon car-body (0.012) Coupe/convertible (0.017) Three doors (0.01) (0.014) (0.039) (0.011) (0.018) (0.004) (0.003) (0.013) (0.013) (0.041) (0.042) (0.011) (0.011) Nb. of observations Notes: Column (1): price not instrumented, (2): evolution of CO 2 preferences captured through a temporal trend. (3): CO 2 interacted with year dummies. (4): CO 2 interacted with three periods ( , and 2008). All specifications include model and years fixed effects. Standard errors are in parentheses. Significance levels: 1%, 5%, 10%. Table 6: Estimates of average preferences for vehicle characteristics. Before studying the way consumers value CO 2, we describe results on their preferences for standard attributes. When not instrumented, the price coefficient has the correct sign but is very small, leading to price elasticities between -1 and 0. Horsepower is also significantly negative in this case. Apart from that, all mean parameters are globally stable from one specification to another. Weight, fuel economy and cylinder capacity are positively valuated. On average, households dislike coupe/convertible, station-wagon cars and threedoor vehicles. 13 We obtain estimates around 0.3 for σ, reflecting the fact that products inside segments are moderate substitutes. This result stems from the fact that in our estimation, we have controlled for model fixed effects. A large part of the correlation on fixed unobservable characteristics of models in the same segment is thus already taken into account. Without model fixed effects, we obtain σ 0.6. All the previous estimates correspond to the average parameters of preferences of pur- 13 Small cars in Europe can have three doors. Such cars correspond to baseline models. 18

21 chasers. Table 19 in Appendix B shows there is a substantial heterogeneity across them. Households are in particular less sensitive to price when they live in urban area than in rural area. The effect of price also decreases with income and age. Hence, unsurprisingly, the more price-sensitive consumers are the young and poor ones, living in rural areas. Using these estimates, we compute the price elasticities for each demographic group, using the fact that in our framework, the price elasticity of product j for group d is equal to β d p j (1 σ s d j/g (1 σ)sd j)/(1 σ)). Sales-weighted average price elasticities are reported in Table 7. These elasticities lie between and -1.54, the mean being These orders of magnitude are similar to those found in the literature. They are below those of Berry et al. (1995), who report price elasticities between -6.5 and -3.5 (see their Table 4) but in the same range as those of Train & Winston (2007), who obtain on average. Moreover, as expected, the most elastic demand comes from young, poor and rural consumers, whereas the less elastic one arises from rich, old and urban individuals. Rural area Urban area Income/Age , (0.361) 3.68 (0.388) 2.72 (0.362) 3.93 (0.348) 3.15 (0.380) 2.24 (0.352) 22,000-32, (0.366) 3.31 (0.384) 2.39 (0,y365) 3.82 (0.367) 2.81 (0.376) 1.97 (0.362) 32, (0.366) 2.80 (0.38) 1.93 (0.367) 3.46 (0.374) 2.35 (0.376) 1.54 (0.369) Notes: the standard errors, in parentheses, are computed by bootstrap. Table 7: Average price elasticity (sales-weighted) according to demographic characteristics. Overall, the previous results on consumers preferences for standard attributes are reassuring and give credit to the model and the estimations. We can thus turn our attention to the estimates of the valuations of CO 2 emissions. 3.3 Evidence of environmental valuation changes Table 8 displays the estimates of the evolution of CO 2 emissions valuation corresponding to Specifications (1) to (4) considered above The parameters presented correspond to the average in the population, parameters of heterogeneity are presented in Table 20 in Appendix B. 19

22 Variable (1) (2) (3) (4) CO 2 emissions (0.038) CO 2 emissions Trend (0.005) (0.042) (0.007) (0.039) (0.039) CO 2 emissions (0.026) CO 2 emissions (0.03) CO 2 emissions (0.022) (0.023) CO 2 emissions ( ) (0.025) Notes: Column (1): price not instrumented, (2): evolution of CO 2 preferences captured through a temporal trend. (3): CO 2 interacted with year dummies. (4): CO 2 interacted with three periods ( , and 2008). 1%, 5%, 10%. Significance levels: Table 8: Estimates of CO 2 emissions valuation (mean parameters) The negative and significant sign of CO 2 emissions captures the idea that consumers have a preference for low CO 2 emitting cars, all things being equal. We also observe a quite clear pattern on evolutions. All specifications indicate a growing concern on CO 2 emissions in purchases. From to 2008, our estimates show that the average of the parameter on CO 2 emissions, which is negatively related to consumers concern on global warming, has decreased by around 0.38 according to Specifications (3) or (4). Moreover, while the trend is quite large, Specification (3) indicates that actually, there have been two main steps in this evolution: a first one in 2006, and a second one in 2008 whereas 2007 resembles much to Column (4), which summarizes these results, indicates that both effects have a similar magnitude with a first decrease of in the years compared to the years and a second decrease of ( ) in 2008 compared to the period Because the compulsory energy label policy took place at the end of 2005 and the feebate was introduced at the beginning of 2008, this timing suggests that the shift in preferences is related to the policies introduced during this period. Up to now, we have estimated preferences for CO 2 emissions. It is possible, however, given that both the energy label and the feebate policies are based on classes of emissions, that consumers focus on such classes rather than directly on CO 2 emissions. This may especially be true after energy labels became compulsory, as the information was more easily transmitted through these labels. To assess the plausibility of this interpretation, we 20

23 estimate a model similar to Specification (4) above, in which CO 2 emissions are replaced by the classes of emissions. Class of emissions Parameter Class of emission Parameter A ( ) (0.179) E+ ( ) (0.028) A (2008) E+ (2008) (0.15) (0.034) B ( ) E- ( ) (0.033) (0.020) B (2008) E- (2008) (0.037) (0.025) C+ ( ) F ( ) (0.03) (0.025) C+ (2008) F (2008) (0.035) C- ( ) (0.027) C- (2008) (0.031) (0.031) G ( ) (0.035) G (2008) (0.043) Notes: the parameters are obtained with the same model as in Column (4) of Table 6, except that we replace CO 2 by the class of emissions dummies. Table 9: Evolution of valuation for classes of emissions (mean parameters) The results, displayed in Table 9, are in line with those on CO 2 emissions. We observe a sharp evolution of consumers preferences towards environmentally-friendly goods during this period. The results are also consistent with the previous interpretation. The raise in the valuation of low-emitting classes (namely, A to C) contrasts with the fall in the highemitting ones (E to F). Only class G, which represents less than 1% of total sales in 2008, has a profile that does not fit with our other results. 15 The raise is especially striking in 2008 for classes A and B. Similarly, the fall for E+ cars is larger in 2008 than in For other polluting cars (classes E- and F), the shift appears to be similar in and in 2008, suggesting that consumers were more attracted by the rebates than discouraged by the fees. In the end, the signals given by these policies, first with colorful labels, then with both labels and prices, seem to have been successful to shift consumers preferences towards environmentally-friendly cars and to align the preferences of the consumers with the classes promoted by the French government. With the previous estimates in hand, we can compute the willingness to pay for a 10g reduction of CO 2 emissions. Because ˆβ is the valuation of price in thousands of euros and 15 The results are consistent with the fact that these cars are luxury goods, with an inelastic demand. 21